Pressure pattern recognition in buildings using an unsupervised machine-learning algorithm

Owing to its significance in ensuring structural safety and occupant comfort, wind pressure on buildings has attracted the attention of numerous scholars. However, the characteristics of wind pressures are usually complex. This study employs an unsupervised machine-learning algorithm, clustering algorithms, to study wind pressures on buildings. Wind pressures on a single building and two adjacent buildings with different gaps are measured in a wind tunnel, with clustering algorithms applied to cluster different wind pressure patterns. The results show that for the single-building model, the pressure patterns are symmetrical on the side surfaces of the building; for the two-building model with a small gap, a channeling effect can be identified; for the two-building model with a large gap, the pressure patterns shared symmetry with that of the single-building model. Clustering algorithms can recognize unidentified patterns of wind pressures on buildings. This study demonstrates that clustering algorithms are a powerful tool for recognizing patterns hidden in complex pressure fields and flow fields. Therefore, this study proposes a promising machine-learning technique that can perfectly complement traditional building methods using wind engineering.